Subgroup analyses of the phase 3 CAIRO3 trial: Association of number and total volume of metastases on survival in patients with metastatic colorectal cancer based on a deep-learning automatic segmentation model.

Abstract

130 Background: Maintenance treatment with capecitabine and bevacizumab after capecitabine, oxaliplatin, and bevacizumab therapy in metastatic colorectal cancer (mCRC) is shown to be effective with observation as an acceptable alternative. Metastatic burden might aid in shared decision-making and the aim of this exploratory analysis is to assess the prognostic and predictive value of number and total volume of metastases in mCRC patients included in the CAIRO3 study. Methods: Patients in the CAIRO3 study with liver and/or lung metastases were included. Every individual liver and lung metastasis was counted and volumes were assessed using deep-learning automatic segmentation software (SAS analytical platform and Philips IntelliSpace). Lymph node metastases were counted manually. Each scan was supervised by a dedicated radiologist. Kaplan-Meier analyses were used to estimate progression-free survival (PFS) and overall survival (OS) using log-rank testing for statistical comparison. Hazard ratios (HR) were estimated using Cox regression models. Results: A total of 156 patients were included with 154 OS events, in which 8198 metastases were identified: 5179 were liver metastases, 2835 lung metastases, and 184 lymph node metastases. Patients had a mean age of 63 years and 67% were male. At randomization, the median number of metastases was 9 and mean total metastatic volume was 116 cm3. Median OS for patients with ≤10 versus >10 metastases was 22.0 months (95%CI: 19.3-29.3) and 14.3 months (95% CI: 11.6-20.9, p-value <0.01), with similar differences for low/high tumor volume. In multivariable analysis number and total volume were independent prognostic markers for PFS and OS, respectively, after adjustment for age, sex, WHO performance score, treatment arm, serum lactate dehydrogenase (LDH) level, site of primary tumor, time to metastases, and treatment response. Benefit of maintenance therapy was most pronounced in patients with ≤10 compared to >10 metastases with a PFS HR of 0.24 (0.12-0.46) compared to 0.52 (0.28-0.96) and OS HR of 0.75 (0.42-1.34) compared to 1.17 (0.66-2.08), demonstrating potential predictive value. Conclusions: Metastatic burden is associated with a poor prognosis in patients with mCRC. Furthermore, patients with limited metastatic disease (1-10 metastases) benefit most from maintenance treatment and this could potentially be a predictive biomarker. [Table: see text]